Integrative computational immunogenomic profiling of cortisol‐secreting adrenocortical carcinoma

Abstract Adrenocortical carcinoma (ACC) is a rare but highly aggressive malignancy. Nearly half of ACC tumours overproduce and secrete adrenal steroids. Excess cortisol secretion, in particular, has been associated with poor prognosis among ACC patients. Furthermore, recent immunotherapy clinical trials have demonstrated significant immunoresistance among cortisol‐secreting ACC (CS‐ACC) patients when compared to their non‐cortisol‐secreting (nonCS‐ACC) counterparts. The immunosuppressive role of excess glucocorticoid therapies and hypersecretion is known; however, the impact of the cortisol hypersecretion on ACC tumour microenvironment (TME), immune expression profiles and immune cell responses remain largely undefined. In this study, we characterized the TME of ACC patients and compared the immunogenomic profiles of nonCS‐ACC and CS‐ACC tumours to assess the impact of differentially expressed genes (DEGs) by utilizing The Cancer Genome Atlas (TCGA) database. Immunogenomic comparison (CS‐ vs. nonCS‐ACC tumour TMEs) demonstrated an immunosuppressive expression profile with a direct impact on patient survival. We identified several primary prognostic indicators and potential targets within ACC tumour immune landscape. Differentially expressed immune genes with prognostic significance provide additional insight into the understanding of potential contributory mechanisms underlying failure of initial immunotherapeutic trials and poor prognosis of patients with CS‐ACC.


| INTRODUC TI ON
Adrenocortical carcinoma (ACC) is among the rarest and most aggressive cancers. Although the current prognostication of patients with ACC primarily hinges on the presence or absence of metastases and tumour resectability, over a third of patients present with an advanced, unresectable disease. [1][2][3][4][5][6] Patients with fully resectable (R 0 ) disease have a reported 5-year survival rate of approximately 50%, whereas patients with the unresectable disease have a 5-year survival rate near 0% and a median survival of shorter than 12 months. 4,7,8 Although large collaborative studies have greatly enhanced the molecular characterization of ACCs 9,10 aside from the advent of mitotane therapy in the treatment of ACC since 1959, there has been little improvement in overall mortality over the past several decades. 11,12 Due to the limited therapeutic options for patients with unresectable ACC, several immunotherapies are currently under evaluation; [13][14][15][16] thus, having a comprehensive understanding of the ACC tumour microenvironment is important for guiding future therapeutic directions.
Nearly half of patients presenting with ACC have been shown to exhibit steroid hormone hypersecretion with excess cortisol secretion being the most predominant hormone and often considered a strong risk factor for poor prognosis. 6,17 Glucocorticoids, including cortisol, are small lipid hormones produced by the adrenal glands that exert their effects through glucocorticoid receptors modulating gene expression to perform a variety of functions, including arresting immune cell growth and maturation, inhibiting activation signalling and inducing lymphocyte apoptosis. 18,19 Glucocorticoids have proven so effective in this role that they are the cornerstone of treatment for many hypersensitive immune reactions and autoimmune diseases. 20,21 However, the immunosuppressive effects of excess glucocorticoid therapy and hypersecretion have also been shown to hinder the immune system's capacity to ward off infections and malignancy and have been associated with a variety of other effects, including muscle wasting, osteoporosis and metabolic derangements. 22,23 A recent in vivo study by Landwehr et al. 24 demonstrated cortisol excess to be associated with T-cell depletion and anergy in ACC TME, while a recent immunotherapy clinical trial revealed a pattern of immune resistance among cortisol-secreting ACC (CS-ACC) tumours, with higher rates of immunotherapeutic failure among CS-ACC patients compared to the patients with nonCS-ACC. 13,[25][26][27] In this study, we utilized The Cancer Genome Atlas (TCGA) ACC cohort 28 to characterize the TME of ACC by comparing TME immunogenomic profiles of CS and nonCS-ACCs. We have also investigated the correlations and prognostic significance of differentially expressed immune genes (DEIGs) and tumourinfiltrating immune cell (TIIC) profiles.

| Data acquisition, patient demographics & tumour pathology
We utilized the RNA sequencing count

| Computational immunogenomic deconvolution
The Cancer Genome Atlas (TCGA, Firehose Legacy) was accessed through cBioPortal (https://www.cbiop ortal.org/). CIBERSORTx was used to estimate tumour-infiltrating immune subsets (including B cells, CD4 + T cells, CD8 + T cells, dendritic cells, macrophages, natural killer cells and neutrophils). CIBERSORTx is a computational immunogenomic platform, a publicly available web-based deconvolution program (https://ciber sortx.stanf ord.edu) 29 . All genes with quantified mRNA expression (log RNA Seq V2 RSEM) in TCGA database (n > 19,000) were compared between CS-and nonCS-ACCs. The significance criteria for DEG were set at a p-value and q-value < 0.05. After characterizing the relationships between DEGs and comparing the expression profiles between CS-ACC and nonCS-ACC TMEs, DEGs were categorized according to biological function using Panther Gene Classification. 30 All DEIGs and TIIC associations were constructed in heatmap format to represent all potential associations and analysed in their relation to patient OS and DFS.
Additionally, the mRNA expression of genes involved in steroid metabolism was analysed for their correlations with TIICs and prognostic DEIGs. Gene expression signatures were compiled by normalizing the sum of the gene mRNA Z-scores (log RNA Seq V2 RSEM) relative to the median on a scale of −5 to 5. Patients with positive cumulative normalized expression levels (≥0.00) were assigned to the high signature expression group and those with negative cumulative normalized expression levels (<0.00) were assigned to the low signature expression group.

| Patient outcomes analysis
Survival analysis was analysed by time-to-event Cox regression models for overall survival (OS) and disease-free survival (DFS). OS was defined as the time from the date of index operation to the date of death. DFS was defined as the time from index operation to the date of documented disease recurrence or death. Kaplan-Meier method and log-rank test were used to compare OS and DFS of ACC patients according to mRNA expression signature profiles. Significance for OS and DFS analysis was set at a p-value < 0.05.

| Statistics
All quantitative comparison, correlation and survival analyses were performed using the 1.1.383 R statistics software (R Core Team Vienna).

| Patient demographic, tumour pathology and treatment parameters of adrenocortical carcinoma
We identified 67 individuals in the TCGA ACC cohort with pathologically confirmed common type ACC with reported mRNA expression data and who did not undergo neoadjuvant therapy. 28 Of these,  Table S1.
Cortisol secretion was not significantly associated with short-

| Differential Immunologic Gene mRNA Expression (DEGs) of cortisol-secreting and noncortisol-secreting adrenocortical carcinoma
Analysis of all genes (n > 19,000) with quantified mRNA expression in TCGA database demonstrated 1,612 differentially expressed genes (DEGs) between CS-and nonCS-ACC tumours. Of these DEGs, 1,021 were classifiable using Panther Genomic Classification ( Figure 1A).
Forty-four (4.3%) genes of those classifiable were identified to be directly related to immunological processes and termed differentially expressed immune genes (DEIGs). On subcategorization of immunological processes using Panther Genomic grouping, DEIGs were primarily involved in immune response and leucocyte activation and maturation. The distribution of immunological processes is summarized in Figure 1B

| Steroid metabolism gene expression comparison and tumour-infiltrating immune cells in cortisol-secreting adrenocortical carcinoma
The mRNA expression of all genes underlying steroid metabolism enzymes, including cortisol synthesis, and influential transcription factors are depicted in Figure 3A  The CCRL2 gene codes for the C-C Motif Chemokine Receptor-Like 2, a non-signalling seven-transmembrane domain receptor related to the atypical chemokine receptor (ACKR) family, however, and its role of this receptor in TME is elusive. ACKRs typically bind chemokines without G protein signalling activation to promote ligand internalization and degradation; however, more importantly, they regulate immune functions by scavenging chemokines from the local environment. 32 Previous studies have demonstrated CCRL2 receptors to act as decoy receptors scavenging chemokines from the TME and their expression to be associated with poor dendritic cell trafficking. 33 Elevated CCRL2 expression has been shown in primary neutrophils relative to other immune cell types and further upregulated on neutrophil activation. 33 Genes with mRNA expression found to be significantly associated

| DISCUSS ION
In this study, we examined and defined the DEIGs and TIIC profiles of ACC tumour microenvironment and immunosuppressive signatures through computational immunogenomic deconvolution of the TCGA genomic data. Specifically, we noted differences between CS-ACC and hormonally inactive or non-cortisol-producing hormonally active ACC tumours (nonCS-ACC). Our findings strongly support previous studies where CS-ACC was shown to be associated with immune resistant TME and poor patient outcomes compared to nonCS-ACC despite similar pathology and stage. 24 Furthermore, we demonstrated immunogenomic differences between CS-ACC and nonCS-ACC TME while identifying distinct mRNA expression profiles associated with immune process genes. The downregulation of many of these DEIGs was associated with poor patient outcomes and differential TIIC profiles. Consistent with a recent independent cohort study, 24 CS-ACC tumours demonstrated significantly lower levels of CD8 T cells compared to nonCS-ACC. Additionally, CS-ACC showed decreased infiltration of NK a cells and M1 macrophages. DC a tumour infiltration was observed to a greater degree in CS-ACC tumours and associated with a poor DFS prognosis. These findings support the notion that excess cortisol secretion in the ACC TME may not only alter the TIIC abundance, diversity and activity, but also contribute to tumour immune escape, immunotherapeutic failure and adversely impact patient outcomes. were primarily related to cellular and metabolic processes and biological regulation; however, a small portion was identified to be directly related to immunological processes (DEIGs). This deductive analysis served as a starting point for our study to further define the potential immunosuppressive role of excess cortisol in the ACC TME.  [42][43][44][45][46] Importantly, however, glucocorticoids have been shown to regulate cytokine secretion in T/NK lymphocytes and potentiate the inhibitory capacity of programmed cell death 1 by upregulating its expression on T cells. 43 In this study, cortisol secretion was associated with decreased CD8 + T and NK a cell infiltration in the TME compared to nonCS-ACC.
This collage of evidence suggests combating glucocorticoid suppression of lymphocytes may serve as a potential therapeutic target worthy of investigation, particularly in CS-ACC tumours.
Excess glucocorticoid signalling has also been shown to inhibit macrophage differentiation towards a proinflammatory phenotype by attenuating the induction of proinflammatory genes that inhibits their differentiation of M1 phenotype. [47][48][49] In our study, we observed significant M2 macrophage infiltration to be the predominant macrophage phenotype in all ACC tumours. Significantly fewer activated M1 macrophages were noted within TME of CS-ACC compared to nonCS-ACC tumours. Increasing macrophage recruitment, maturation and activation may be another means of TME optimization and a potential avenue for future therapeutic development in CS-ACCs.
Stimulation of the glucocorticoid receptor impacts NF-κB family proteins to inhibit their transcriptional activity. 50 This results in innate and adaptive immune suppression through the decreased expression of co-stimulatory molecules, cytokines and chemokines as well as the upregulation of co-inhibitory molecules. 49,50 As observed in this study, several downstream NF-κB product genes showed downregulation in CS-ACC (CCR2, CD40, CD40LG, EOMES, TLR5). 48,49,51 Additionally, glucocorticoid-mediated inhibition of NF-κB signalling pathways has been shown to hinder DC maturation and antigen presentation efficiency. [51][52][53] Although the use of the TCGA database empowered this study by providing a sufficiently robust database of clinicogenomic parameters to derive meaningful associations in characterizing the TME of these ultra-rare tumours, the collaborative is limited to large, academic referral centres which may lead to selection bias towards more aggressive, later stage disease with over-representation of CS-ACC and metastatic disease. This potential selection bias may limit the generalizability of our conclusions. Furthermore, the collaborative nature of the TCGA database also limits the granularity of clinical data available. For example, the TCGA database only reports on ACC hormone hypersecretion (nonfunctional, cortisol, aldosterone, oestrogen etc.) and does not include the diagnostic test use or laboratory values. Our analysis was limited to utilizing clinical and biochemical evaluation of excess cortisol production as a surrogate for degree of Cushing's disease which showed a trend towards more severe immune suppressive immunogenomic and TIIC profiles but was limited by the low statistical power of sub-analysis ( Figure S1). Furthermore, the treatment of patients with ACC is very heterogeneous across institutions with variations in surgical technique, radiation therapy and mitotane regimen (including dose, frequency and therapeutic level). Altogether, such limitations hinder our ability to further characterize and account for many clinical and treatment factors that may impact OS and DFS in CS-and nonCS-ACC patients. Additionally, our analysis is limited to bulk sample mRNA sample deconvolution using the CIBERSORTx algorithm.
Additionally, although the deductive design of this study benefits the sensitivity for identifying DEIGs between CS-and nonCS-ACCs, this method, along with a relatively small patient population, may limit the specificity of our analysis, thus increasing the potential for type 2 errors and false positive correlations. Similarly, although there were no statistically significant differences in demographics, treatment and tumour stage/pathology between CS-and nonCS-ACC patient groups, it is plausible that accumulation of factors more prevalent in the CS-ACC group-but not statistically different-may conspire to negatively impact survival, potentially confounding the correlations identified in this study.
The CIBERSORTx algorithm for TIIC estimation is highly correlative for certain immune cell populations, including CD8 T cells and Bcell subtypes; however, these methods are less precise at estimating DC populations and DC subtypes. 54 Nonetheless, DC estimations were included in our analysis due to their crucial role in potentiating ICI and T-cell activation. Although DC a infiltration was increased in CS-compared to nonCS-ACC and associated with shortened OS and DFS, we suspect a molecular process of DCs may be influenced by excess cortisol in the ACC TME that we are unable to investigate further with the available data (such as DC migration, maturation and antigen presentation efficiency). Mature activated DCs are equipped to capture antigens and to produce large numbers of immunogenic MHC-peptide complexes to potentiate T-cell immunity. However, glucocorticoids have been shown to distinctly alter the phenotype of DCs by stunting maturation, hindering migration and inhibiting the expression of MHC proteins. [51][52][53] The overall impact of glucocorticoids on DCs has been summarized as a partial conversion to a monocyte-macrophage phenotype and impaired capacity to reach maturation resulting in decreased T-cell stimulation. 51,52 Altogether the impact of excess cortisol among CS-ACC may result in increased accumulation of inefficient DC a in the ACC TME.
This study may offer additional insight into why strong immune infiltration is rarely seen in CS-ACC and why current immunological therapeutic options have been of limited efficacy. Our findings suggest that the ACC cortisol hypersecretion impacts TME in favour of immune resistance. Excess cortisol in the ACC TME may potentially facilitate more aggressive tumour biology and poor prognosis. To date, several studies have now highlighted the negative effects of synthetic glucocorticoids on the outcome of immunotherapy. 41,42 In line with this, patients with CS-ACC were recently reported to experience decreased response to immunotherapy and experience poor patient outcomes compared to nonCS-ACC patients when treated with anti-PD-L1 agent Pembrolizumab and mitotane. 25 This study findings support two previous related but different Furthermore, the mRNA expression levels of several genes coding for cortisol synthesis enzymes (CYP11A1, CYP17A1, CYP21A2) and steroid metabolism transcription factors (PDX1, NR5A1) were upregulated in CS-ACC and associated with decreased CD8 + T-cell infiltration. These gene products and pathways may provide for actionable drug targets to combat immune resistance in CS-ACCs.
In summary, our study characterized a distinct immunogenomic profile with a significant prognostic value associated with CS-ACC compared to nonCS-ACC that may contribute to the poor outcomes associated in patients with CS-ACC. In depth future studies aimed at uncovering the full impact of excess glucocorticoid metabolism and secretion in TME and comprehensive targeting of steroid metabolism may provide new immunotherapeutic applications for effective treatment of aggressive and poorly responsive CS-ACC tumours to improve patient survival. Our findings may help guide future studies needed to clarify the potential mechanisms of immune resistance and immunotherapy failure in CS-ACC. Such insight may empower strategies to reduce the potentially harmful effects of excess cortisol secretion and synthetic glucocorticoids used to control side effects and symptoms associated with many immunotherapies.